duan8/scaled-yolov4/common.hpp

#include <fstream>
#include <map>
#include <sstream>
#include <vector>
#include <opencv2/opencv.hpp>

#include "NvInfer.h"
#include "yololayer.h"
#include "mish.h"


using namespace nvinfer1;

cv::Mat preprocess_img(cv::Mat& img) {
    int w, h, x, y;
    float r_w = Yolo::INPUT_W / (img.cols*1.0);
    float r_h = Yolo::INPUT_H / (img.rows*1.0);
    if (r_h > r_w) {
        w = Yolo::INPUT_W;
        h = r_w * img.rows;
        x = 0;
        y = (Yolo::INPUT_H - h) / 2;
    } else {
        w = r_h* img.cols;
        h = Yolo::INPUT_H;
        x = (Yolo::INPUT_W - w) / 2;
        y = 0;
    }
    cv::Mat re(h, w, CV_8UC3);
    cv::resize(img, re, re.size());
    cv::Mat out(Yolo::INPUT_H, Yolo::INPUT_W, CV_8UC3, cv::Scalar(128, 128, 128));
    re.copyTo(out(cv::Rect(x, y, re.cols, re.rows)));
    return out;
}

cv::Rect get_rect(cv::Mat& img, float bbox[4]) {
    int l, r, t, b;
    float r_w = Yolo::INPUT_W / (img.cols * 1.0);
    float r_h = Yolo::INPUT_H / (img.rows * 1.0);
    if (r_h > r_w) {
        l = bbox[0] - bbox[2]/2.f;
        r = bbox[0] + bbox[2]/2.f;
        t = bbox[1] - bbox[3]/2.f - (Yolo::INPUT_H - r_w * img.rows) / 2;
        b = bbox[1] + bbox[3]/2.f - (Yolo::INPUT_H - r_w * img.rows) / 2;
        l = l / r_w;
        r = r / r_w;
        t = t / r_w;
        b = b / r_w;
    } else {
        l = bbox[0] - bbox[2]/2.f - (Yolo::INPUT_W - r_h * img.cols) / 2;
        r = bbox[0] + bbox[2]/2.f - (Yolo::INPUT_W - r_h * img.cols) / 2;
        t = bbox[1] - bbox[3]/2.f;
        b = bbox[1] + bbox[3]/2.f;
        l = l / r_h;
        r = r / r_h;
        t = t / r_h;
        b = b / r_h;
    }
    return cv::Rect(l, t, r-l, b-t);
}

float iou(float lbox[4], float rbox[4]) {
    float interBox[] = {
        std::max(lbox[0] - lbox[2]/2.f , rbox[0] - rbox[2]/2.f), //left
        std::min(lbox[0] + lbox[2]/2.f , rbox[0] + rbox[2]/2.f), //right
        std::max(lbox[1] - lbox[3]/2.f , rbox[1] - rbox[3]/2.f), //top
        std::min(lbox[1] + lbox[3]/2.f , rbox[1] + rbox[3]/2.f), //bottom
    };

    if(interBox[2] > interBox[3] || interBox[0] > interBox[1])
        return 0.0f;

    float interBoxS =(interBox[1]-interBox[0])*(interBox[3]-interBox[2]);
    return interBoxS/(lbox[2]*lbox[3] + rbox[2]*rbox[3] -interBoxS);
}

bool cmp(const Yolo::Detection& a, const Yolo::Detection& b) {
    return a.det_confidence > b.det_confidence;
}

void nms(std::vector<Yolo::Detection>& res, float *output, float conf_thresh, float nms_thresh = 0.5) {
    int det_size = sizeof(Yolo::Detection) / sizeof(float);
    std::map<float, std::vector<Yolo::Detection>> m;
    for (int i = 0; i < output[0] && i < Yolo::MAX_OUTPUT_BBOX_COUNT; i++) {
        if (output[1 + det_size * i + 4] <= conf_thresh) continue;
        Yolo::Detection det;
        memcpy(&det, &output[1 + det_size * i], det_size * sizeof(float));
        if (m.count(det.class_id) == 0) m.emplace(det.class_id, std::vector<Yolo::Detection>());
        m[det.class_id].push_back(det);
    }
    for (auto it = m.begin(); it != m.end(); it++) {
        //std::cout << it->second[0].class_id << " --- " << std::endl;
        auto& dets = it->second;
        std::sort(dets.begin(), dets.end(), cmp);
        for (size_t m = 0; m < dets.size(); ++m) {
            auto& item = dets[m];
            res.push_back(item);
            for (size_t n = m + 1; n < dets.size(); ++n) {
                if (iou(item.bbox, dets[n].bbox) > nms_thresh) {
                    dets.erase(dets.begin()+n);
                    --n;
                }
            }
        }
    }
}

// TensorRT weight files have a simple space delimited format:
// [type] [size] <data x size in hex>
std::map<std::string, Weights> loadWeights(const std::string file) {
    std::cout << "Loading weights: " << file << std::endl;
    std::map<std::string, Weights> weightMap;

    // Open weights file
    std::ifstream input(file);
    assert(input.is_open() && "Unable to load weight file.");

    // Read number of weight blobs
    int32_t count;
    input >> count;
    assert(count > 0 && "Invalid weight map file.");

    while (count--)
    {
        Weights wt{DataType::kFLOAT, nullptr, 0};
        uint32_t size;

        // Read name and type of blob
        std::string name;
        input >> name >> std::dec >> size;
        wt.type = DataType::kFLOAT;

        // Load blob
        uint32_t* val = reinterpret_cast<uint32_t*>(malloc(sizeof(val) * size));
        for (uint32_t x = 0, y = size; x < y; ++x)
        {
            input >> std::hex >> val[x];
        }
        wt.values = val;
        
        wt.count = size;
        weightMap[name] = wt;
    }

    return weightMap;
}

IScaleLayer* addBatchNorm2d(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, std::string lname, float eps) {
    float *gamma = (float*)weightMap[lname + ".weight"].values;
    float *beta = (float*)weightMap[lname + ".bias"].values;
    float *mean = (float*)weightMap[lname + ".running_mean"].values;
    float *var = (float*)weightMap[lname + ".running_var"].values;
    int len = weightMap[lname + ".running_var"].count;

    float *scval = reinterpret_cast<float*>(malloc(sizeof(float) * len));
    for (int i = 0; i < len; i++) {
        scval[i] = gamma[i] / sqrt(var[i] + eps);
    }
    Weights scale{DataType::kFLOAT, scval, len};
    
    float *shval = reinterpret_cast<float*>(malloc(sizeof(float) * len));
    for (int i = 0; i < len; i++) {
        shval[i] = beta[i] - mean[i] * gamma[i] / sqrt(var[i] + eps);
    }
    Weights shift{DataType::kFLOAT, shval, len};

    float *pval = reinterpret_cast<float*>(malloc(sizeof(float) * len));
    for (int i = 0; i < len; i++) {
        pval[i] = 1.0;
    }
    Weights power{DataType::kFLOAT, pval, len};

    weightMap[lname + ".scale"] = scale;
    weightMap[lname + ".shift"] = shift;
    weightMap[lname + ".power"] = power;
    IScaleLayer* scale_1 = network->addScale(input, ScaleMode::kCHANNEL, shift, scale, power);
    assert(scale_1);
    return scale_1;
}

ILayer* convBnMish(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int outch, int ksize, int s, int p, int linx) {
    Weights emptywts{DataType::kFLOAT, nullptr, 0};
    IConvolutionLayer* conv1 = network->addConvolutionNd(input, outch, DimsHW{ksize, ksize}, weightMap["module_list." + std::to_string(linx) + ".Conv2d.weight"], emptywts);
    assert(conv1);
    conv1->setStrideNd(DimsHW{s, s});
    conv1->setPaddingNd(DimsHW{p, p});

    IScaleLayer* bn1 = addBatchNorm2d(network, weightMap, *conv1->getOutput(0), "module_list." + std::to_string(linx) + ".BatchNorm2d", 1e-4);

    auto creator = getPluginRegistry()->getPluginCreator("Mish_TRT", "1");
    const PluginFieldCollection* pluginData = creator->getFieldNames();
    IPluginV2 *pluginObj = creator->createPlugin(("mish" + std::to_string(linx)).c_str(), pluginData);
    ITensor* inputTensors[] = {bn1->getOutput(0)};
    auto mish = network->addPluginV2(&inputTensors[0], 1, *pluginObj);
    return mish;
}